A method for controlling diseases and pests in cultivation of dendrobium

Abstract

The application provides a disease and pest control method for Dendrobium cultivation, comprising: obtaining a UAV aerial image, identifying a suspected disease and pest area based on the UAV aerial image, and sending the position of the suspected disease and pest area to a background system; obtaining environmental parameters, analyzing the influence weight of the environmental parameters on diseases and pests, and performing disease and pest risk prediction to obtain a risk prediction result; starting a control measure when the risk value exceeds a warning threshold according to the risk prediction result, adjusting the environment through a sunshade and a ventilation device, collecting a plant image to identify a disease spot coverage rate, and adjusting a control frequency according to the coverage rate.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for controlling pests and diseases in Dendrobium cultivation. Background Technology

[0002] A key technical challenge in Dendrobium cultivation is the precise monitoring and control of environmental parameters. Dendrobium growth is highly sensitive to environmental factors such as temperature, humidity, and light; even slight fluctuations can negatively impact its growth, development, yield, and quality. Currently, most Dendrobium cultivation relies on manual experience to control the environment, lacking real-time, dynamic data monitoring and feedback mechanisms, making it difficult to achieve optimal growth conditions. Furthermore, different growth stages and varieties of Dendrobium have varying environmental requirements, and existing environmental control methods are relatively crude and cannot meet the dynamic needs of Dendrobium growth. Therefore, there is an urgent need to establish a precise environmental monitoring and control system. This system should use sensors to collect real-time data on temperature, humidity, and light, and combine this data with Dendrobium physiological characteristic models to dynamically adjust environmental parameters, creating optimal growth conditions for Dendrobium at different growth stages and improving both yield and quality. Summary of the Invention

[0003] This invention provides a method for controlling pests and diseases in Dendrobium cultivation, mainly including: The system acquires aerial images taken by drones, identifies suspected pest and disease areas based on these images, and sends the locations of the suspected pest and disease areas to the backend system. The back-end system controls the drone to perform a detailed scan of the suspected pest and disease area, and combines color difference thermal imaging detection to confirm the type and distribution of pests and diseases. Based on the types and distribution of pests and diseases, analyze the migration paths of pests, calculate the release parameters of natural enemy insects, and generate a map of ecological barrier points in ditches using the drone aerial images to obtain a prevention and control plan. According to the prevention and control plan, natural enemy insects are released through automatic release equipment, and water level sensors are used to monitor the water level in the ditch. When the water level is lower than the threshold, water replenishment is initiated, and the barrier effect is evaluated by the barrier rate. Environmental parameters are acquired, their impact weights on pests and diseases are analyzed, and pest and disease risk is predicted to obtain the risk prediction results. Based on the risk prediction results, when the risk value exceeds the warning threshold, control measures are initiated, and plant images are collected to identify the lesion coverage rate. The control frequency is then adjusted based on the lesion coverage rate. The system acquires aerial images of Dendrobium plants taken by drones and uses time series analysis to fit growth curves. When growth indicators continuously deviate from the normal range, it sends early warning information. Monitor the water quality of the ditch and obtain data from the measuring points in the cultivation area. Start the circulation filtration and water replenishment. Perform spatial analysis on the measuring point data in the cultivation area to determine the control parameters of the shading and ventilation devices and maintain the optimal environment for Dendrobium growth.

[0004] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method for controlling pests and diseases in Dendrobium cultivation. The method utilizes a drone equipped with a high-resolution camera and multispectral sensors to regularly inspect the Dendrobium cultivation area, acquiring image data and analyzing it to identify areas suspected of being infested by pests or diseases. For areas identified as at risk, the invention automatically generates control plans, including designing ecological barriers in ditches and releasing natural enemy insects. The invention also establishes a correlation model between environmental factors and pests and diseases, which can predict the probability and severity of pest and disease occurrence and automatically generate early warning information. When the early warning level reaches a threshold, the invention automatically initiates control measures. Furthermore, the invention uses time-series analysis to assess the growth status of Dendrobium and promptly issues warnings when abnormalities are detected. Through intelligent monitoring and analysis, this invention achieves accurate identification, timely early warning, and automatic control of Dendrobium pests and diseases, significantly improving control efficiency and effectiveness, and providing a comprehensive intelligent solution for Dendrobium cultivation. Attached Figure Description

[0005] Figure 1 This is a flowchart of a method for controlling pests and diseases in Dendrobium cultivation according to the present invention.

[0006] Figure 2 This is a schematic diagram of a method for controlling pests and diseases in Dendrobium cultivation according to the present invention.

[0007] Figure 3 This is another schematic diagram of a method for controlling pests and diseases in Dendrobium cultivation according to the present invention. Detailed Implementation

[0008] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0009] like Figure 1-3 This embodiment of a method for controlling pests and diseases in Dendrobium cultivation may specifically include: S101. Regularly inspect the Dendrobium cultivation area using drones equipped with high-resolution cameras and multispectral sensors to acquire Dendrobium image data. Analyze and process the acquired image data; if pest or disease characteristics are identified, the area is identified as a suspected pest or disease infestation area, and the location information of that area is sent to the backend system.

[0010] Multispectral image data, including raw images in the red and near-infrared bands, is acquired using a drone cruiser. The raw images are then denoised using Gaussian denoising to obtain a clear image. This clear image is then divided into pixel blocks by an image segmenter, and plant leaf reflectance data is extracted. Normalization is performed on the plant leaf reflectance data to obtain a plant leaf health index. If the health index is less than a threshold, the latitude and longitude data of the suspected pest / disease infestation area are acquired using a GPS device. Spectral reflectance curves are extracted from the suspected pest / disease infestation area and matched against a pest / disease characteristic spectral database using a spectral matcher. If the similarity is greater than a similarity threshold, the area is identified as an actual pest / disease infestation area, and the area monitoring record is transmitted to the backend database via a wireless data transmitter.

[0011] Specifically, the drone intelligent cruiser cruises sequentially along a fixed route in the Dendrobium cultivation area. Based on feedback data from the land flatness sensor, it automatically adjusts its flight altitude to a range of 8 to 12 meters. It acquires raw image data containing red and near-infrared bands from a multispectral camera and stores the raw image data on the onboard solid-state drive.

[0012] An image denoising algorithm was used to perform Gaussian denoising on the original image data to obtain a clear image. An image segmenter then divided the clear image into several pixel blocks. Leaf reflectance data was extracted from these pixel blocks. The reflectance data in the red and near-infrared bands were normalized to obtain the plant leaf health index. The formula used is: in, This is an index of plant leaf health. For near-infrared band (850nm) reflectance, This refers to the reflectance in the red light band (760nm). Healthy leaves, due to their high chlorophyll content, have strong absorption and low reflectance (approximately 0.1) in the red light band and high reflectance (approximately 0.6) in the near-infrared band, resulting in a higher calculated H value. Conversely, in damaged leaves, chlorophyll degradation leads to a lower H value. An H value below 0.4 indicates a suspected pest or disease infestation area.

[0013] Support vector machine is used to perform binary classification on the plant leaf health index. Standard values ​​are read from the plant leaf feature database. If the current leaf health index is less than 90% of the standard value, it is identified as a suspected pest or disease area. Global positioning equipment is used to obtain the latitude and longitude data of the area and generate a region identifier and latitude and longitude mapping record.

[0014] Spectral reflectance curves are extracted from suspected pest and disease areas based on the mapping records. These curves are then matched with a pest and disease characteristic spectral database using a spectral matcher. If the similarity is greater than 0.85, the area is identified as an actual pest and disease area. The coordinates of the center point of the pest and disease area are calculated, and the monitoring record containing the area identifier, latitude and longitude coordinates, and spectral reflectance curve is sent to the backend database by a wireless data transmitter.

[0015] The drone patrol routes in the Dendrobium cultivation area are typically arranged perpendicular to the planting rows, with a single flight path spacing of 8 meters. This matches the field of view of the high-definition lens, ensuring an image overlap rate of 60% and avoiding blind spots in image stitching. The land levelness sensor uses ultrasonic ranging, emitting ultrasonic waves to the ground and receiving reflected signals. Image resolution decreases at excessively high altitudes, while the field of view narrows at excessively low altitudes; therefore, a reasonable flight altitude range of 8 to 12 meters is set. A multispectral camera simultaneously acquires reflectance data in both the 760nm red band and the 850nm near-infrared band, recording the reflection characteristics of plant leaves to different wavelengths of light. Gaussian denoising processing applies a 5×5 pixel Gaussian kernel to the original images, effectively eliminating random noise generated during image acquisition.

[0016] The image segmenter divides the clear image into 32×32 pixel blocks, with each block roughly corresponding to a 0.5 square meter area on the actual ground. The plant leaf health index is obtained by calculating the Normalized Difference Vegetation Index (NDVI). Healthy plant leaves have high reflectivity in the near-infrared band (above 0.6) and low reflectivity in the red band (around 0.1), indicating sufficient chlorophyll content. A support vector machine (SVM) constructs a classification hyperplane using radial basis function kernels to perform binary classification on the plant leaf health index. By comparing the index with 3000 sets of healthy leaf data stored in the plant leaf feature database, leaves with a health index below 90% of the standard value are classified as suspected pests or diseases.

[0017] The global positioning device uses BeiDou satellite positioning, with a positioning accuracy better than 1 meter, and records the latitude and longitude information of suspected pest and disease areas. The spectral matcher uses the correlation coefficient method to calculate the similarity between the measured spectrum and the spectra of known pests and diseases in a characteristic spectral library. The characteristic spectral library contains standard spectral curves for 20 common pests and diseases. Leaves damaged by spider mites show a significant reduction in reflectance in the 750 nm to 850 nm wavelength band, only 70% of that of healthy leaves. Leaves damaged by anthracnose show a distinct reflectance peak in the 680 nm to 720 nm wavelength band, with reflectance 25% higher than healthy leaves. Each spectral curve contains 512 wavelength points in the 400 nm to 1000 nm wavelength band. Regional monitoring records are transmitted in real-time to the backend database via a 4G wireless network, with each record containing approximately 50KB of data.

[0018] S102. After receiving the location information of the suspected pest and disease area, the backend system controls the drone to carry color difference and thermal imaging equipment to perform a fine scan of the area, detect the physiological changes of the Dendrobium plants, and determine whether there is a disease based on the preset disease characteristic threshold. If the characteristics of the disease are detected, it is determined that there is a risk of disease in the area.

[0019] It should be noted that color difference thermal imaging detection refers to the comprehensive identification of pests and diseases by simultaneously utilizing color differences (color difference) in visible light images and temperature differences in thermal infrared images. Color difference detection is based on the color difference values ​​of adjacent pixels in the grayscale histogram of plant leaf areas to detect abnormal spots on the leaf surface; thermal imaging detection acquires temperature distribution data of plant leaves and roots through thermal imaging sensors and uses support vector machines to analyze the degree of temperature anomalies. The results of the two detections corroborate each other. When color differences frequently overlap with areas of temperature anomalies, these areas are identified as suspected pest and disease infestations, thereby improving the accuracy and anti-interference capability of the identification.

[0020] Support Vector Machines (SVMs) are supervised learning models used for binary classification tasks. Their basic architecture involves finding an optimal hyperplane in a high-dimensional feature space that maximizes the margin between samples of different categories (normal temperature and abnormal temperature). Specifically, temperature values ​​from multiple sampling points on plant leaves and roots are used to construct feature vectors. These vectors are mapped to a high-dimensional space using a radial basis function (RBF) kernel, and a classification decision function is trained to output the probability value of the degree of temperature anomaly. In this method, the input to the SVM is the measured values ​​from 16 temperature sampling points, and the output is a temperature anomaly classification label (0 represents normal, 1 represents abnormal). When the output is 1, it is determined to be a suspected disease point.

[0021] This embodiment uses the ResNet network (Residual Neural Network), a deep convolutional neural network. Its core architecture introduces skip connections, allowing the input signal to bypass one or more convolutional layers and be directly transmitted to the output, thereby solving the gradient vanishing problem in deep networks. In this method, the ResNet network can be composed of multiple stacked residual blocks. Each residual block contains two convolutional layers (each with a 3×3 kernel), a batch normalization layer, and a ReLU activation function. The network input consists of the color difference feature distribution map and temperature feature distribution map of suspected disease points (merged into a multi-channel image). After passing through several residual blocks, high-dimensional feature vectors are extracted. Finally, the feature distribution map is output through a fully connected layer for similarity matching with standard feature templates in the disease feature database.

[0022] The system receives spectral sensor data and thermal imaging sensor data collected by a drone. This data is used to scan suspected pest and disease areas according to a sampling grid. Based on the data, region segmentation is performed to obtain a grayscale histogram of the plant leaf area. If the color difference between adjacent pixels in the grayscale histogram exceeds a threshold, an abnormal spot map is obtained. A temperature data map identifier is used to sample the temperature of the plant leaves and roots. A support vector machine is used to calculate the degree of temperature anomaly. If the degree of temperature anomaly exceeds a preset temperature threshold, the area is determined to be a suspected pest or disease point. The degree of temperature anomaly is calculated using the following formula: in, This is the temperature feature vector formed by 16 temperature sampling points on the plant's leaves and roots. For the first Support vectors, Category labels (+1 indicates normal, -1 indicates abnormal). For Lagrange multipliers, The parameter for the radial basis function is 0.5. This is a bias term. Determined to be normal. It was identified as a suspected disease site.

[0023] For the suspected disease points, a ResNet network is used to extract feature distribution maps, and similarity matching is performed using standard feature templates in the disease feature library. If the similarity exceeds a preset matching threshold, it is determined that the corresponding disease exists in the area.

[0024] Specifically, the latitude and longitude coordinates of suspected pest and disease areas are retrieved from the background database. Based on the coordinate data, inspection instructions are issued to the drone. The drone scans the area according to a 0.5-meter by 0.5-meter sampling grid. The surface reflectance data of plant leaves is collected by a spectral sensor in the 400 to 800 nanometer band, and the temperature distribution data of the plant is recorded by a thermal imaging sensor with a sampling frequency of 30 Hz. A digital filter is used to remove noise from the collected spectral and temperature data.

[0025] A region segmenter was used to divide the denoised spectral data into regions, and grayscale histograms of the plant leaf regions were extracted. The color difference value of the leaf surface was calculated based on the grayscale distribution of the RGB channels. If the color difference value between adjacent pixels on a single leaf surface exceeded 0.3, it was recorded as an abnormal spot. The number and area distribution of spots were statistically analyzed from the abnormal spot map. A temperature data map identifier was used to select 16 temperature sampling points in both the plant leaf and root regions. The temperature data of normal plants within a 0.5-meter radius of the plant were compared. The degree of temperature anomaly in the plant leaves and roots was calculated using a support vector machine. If the leaf temperature deviation exceeded 3 degrees Celsius or the root temperature deviation exceeded 4 degrees Celsius, the area was marked as a suspected disease point. A disease feature identifier was used to extract features from suspected disease points. Based on the ResNet network, features were extracted from the color difference feature distribution map and temperature feature distribution map of plant leaves. Standard feature templates for leaf spot disease and root rot disease were read from the disease feature database. Similarity matching was performed on the extracted features. If the color difference feature similarity exceeded 0.85 and the temperature feature similarity exceeded 0.8, the corresponding disease was determined to exist in the area.

[0026] In the precise inspection using drones, the sampling grid was set based on the average size of Dendrobium leaves, with a single leaf typically ranging from 4 to 6 centimeters in length. Therefore, a 0.5-meter by 0.5-meter sampling grid was chosen to ensure sampling accuracy while avoiding data redundancy. A spectral sensor collected the leaf reflectance spectrum in the 400-800 nanometer band, with the green band showing the strongest reflectance at 500-600 nanometers. Brown spots caused by leaf spot disease exhibit a significant decrease in reflectance in this band. Healthy leaves show a reflectance of 0.4 at 550 nanometers, while diseased areas show only 0.15. A thermal imaging sensor with a sampling frequency of 30 Hz was selected, corresponding to the drone's flight speed of 0.5 meters per second, ensuring that the spatial distance between adjacent sampling points was within 2 centimeters. The color difference value of the leaf surface was calculated using the grayscale differences of the three RGB channels. The color difference value of healthy leaves is typically within 0.1, while in the presence of diseased areas, the color difference value in localized areas increases significantly.

[0027] Actual measurements show that in the early stages of leaf spot disease, the spots are 2 to 3 mm in diameter, with a color difference value of 0.35 between adjacent pixels. As the disease progresses, the spots gradually enlarge and merge, and the color difference value further increases to over 0.5. Root rot causes water loss in the leaves, resulting in a uniform fading of the entire leaf surface. At this stage, the overall color difference value is small, but the difference from healthy leaves is significant. The assessment of the degree of abnormal plant temperature is based on local temperature comparison. Sixteen evenly distributed sampling points were selected in the leaf area to construct a temperature feature vector. The temperature difference between the leaf temperature of a normally growing Dendrobium plant and the surrounding air temperature is within 1 degree Celsius. Leaf spot disease causes local tissue necrosis in the leaves, affecting transpiration and causing the temperature at the lesion site to rise by 3 to 4 degrees Celsius. Root rot reduces the water absorption capacity of the roots, leading to an increase in the temperature of all leaves. Furthermore, the rotten areas of the roots experience an abnormal temperature rise of 4 to 5 degrees Celsius due to tissue decomposition.

[0028] Disease feature identification employs deep learning methods to extract disease features. Color difference features mainly include information such as the shape, size, and distribution density of spots. A typical characteristic of leaf spot disease is round or oval brown spots, 2 to 8 mm in diameter, with clear edges. Temperature features include information such as the range, degree, and distribution pattern of temperature anomalies. A typical characteristic of root rot disease is the appearance of a band-like warming area on the roots, 5 to 10 mm wide, with the temperature anomaly lasting for more than 24 hours. Feature similarity is calculated using cosine distance, with empirical thresholds set at 0.85 and 0.8 respectively. Suspected diseases below these thresholds are marked as objects to be observed.

[0029] S103. For areas identified as having pest and disease risks, corresponding prevention and control plans are generated. A neural network algorithm is used to analyze pest migration paths and soil permeability data, comparing these with aquatic ecosystem parameter data to generate a map of ecological barrier locations in ditches. Based on the pest and disease risk level and the coordinates of the risk area, combined with the activity range of natural enemy insects, a random forest method is used to calculate the release quantity parameters and release point coordinates of natural enemy insects.

[0030] Based on the pest density distribution data obtained from the collection grid, a convolutional neural network is used to track the pest activity trajectory to obtain a pest migration path trajectory map; soil moisture sensors are used to obtain soil permeability values, and the location of the barrier zone is determined based on the spatial overlap between the pest migration path and the permeability values; monitoring units are divided according to the monitoring grid, and the risk level of each unit is obtained through an insect density monitor; the number of natural enemy insects to be released is determined based on the predation parameters of natural enemy insects; if the number of natural enemy insects to be released exists in the monitoring unit, a random forest algorithm is used to optimize the spatial distribution of the number of natural enemy insects, the coordinates of the release point are calculated based on the predation radius, and the number of natural enemies at the release point is determined according to the risk level.

[0031] Specifically, the system reads the boundary coordinates of the pest-affected areas from the database, acquires pest density distribution data at four time points daily through an image acquisition grid with a fixed spacing of 1 meter, tracks the pest activity trajectory over 180 consecutive days using a convolutional neural network to generate a pest migration path trajectory map, collects 24-hour soil moisture content and permeability values ​​using a soil moisture sensor buried at a depth of 20 cm, and calculates the optimal barrier zone location based on the spatial overlap between the pest migration path and permeability.

[0032] Based on the soil permeability distribution map of the pest-affected area and the location of the barrier zone, a depth sounder was used to measure the soil profile moisture content of the barrier zone. An ecological barrier was delineated 5 meters outward from the boundary of the pest activity range. The spacing between barrier installations was calculated using an ecological barrier locator. Monitoring points were set up every 10 meters to collect moisture permeability data, generating an ecological barrier layout map including geographic coordinates and installation spacing. A pest density monitor was used to divide the pest-affected area into monitoring units using a 10m x 10m grid. The risk level was determined based on the pest density values ​​within each monitoring unit. Predation radius and predation efficiency parameters of natural enemy insects were extracted from the pest population database. A data fusion device was used to overlay the risk level distribution map with the natural enemy insect predation range map to calculate the required number of natural enemy insects to be released for each monitoring unit.

[0033] Random forests were used to optimize the spatial distribution of predator insect populations within a single grid cell. The effective control area for a single predator insect was calculated based on a 5-meter predation radius. The minimum distance between release points was calculated from the area of ​​the monitoring unit and the required number of predators. Coordinates of predator release points were generated using a hexagonal honeycomb layout. The number of predators at each release point was dynamically allocated based on the risk level of surrounding units. High-precision image acquisition grids were used to track pest migration paths. Density data was collected at four time points: 4 AM, 10 AM, 4 PM, and 10 PM. These time points were chosen based on pest activity patterns; for example, aphids migrate more frequently in the early morning and evening when temperatures are lower, while their activity decreases during the midday heat.

[0034] The 180-day continuous observation period covered the pest's breeding cycle and seasonal changes. Records showed that during peak breeding season, the density of pests at a single monitoring point could reach 800 individuals per square meter, with a migration and spread rate of 2.5 meters per day. Soil moisture sensors were buried at a depth of 20 cm, primarily considering the root growth characteristics of Dendrobium officinale. The top 15-25 cm layer is the area with the densest root distribution and is also the key area for pest and disease control. Soil permeability was measured continuously for 24 hours, recording changes in infiltration after rain. The permeability of healthy soil was 50-80 mm / hour, while in severely affected areas, due to root damage, the permeability dropped to below 20 mm / hour. The ecological barrier was constructed using a grid layout of monitoring points, with one monitoring point every 10 meters to form a uniformly distributed monitoring network. Monitoring data showed that when soil moisture content was maintained at 25%-30%, the pest spread rate significantly decreased, with the movement distance shortened to 0.8 meters per day. The ecological barrier area extended outwards by 5 meters, and drip irrigation lines were installed every 2 meters within the barrier zone to maintain appropriate soil moisture content. The predatory insects are released in a honeycomb hexagonal layout, with each hexagon having a side length of 5 meters, which matches the effective predatory radius of the predatory insects.

[0035] Taking ladybugs as an example, adult ladybugs prey on 80 to 100 aphids daily, with an activity radius of 4 to 6 meters. In areas with high pest infestations, 2 to 3 ladybugs are released per square meter. Data shows that when the density of natural enemy insects reaches 5% of the pest density, the pest population begins to decline significantly. The selection of release sites considers terrain and vegetation distribution, increasing the release density in areas with dense vegetation to compensate for the reduced predation efficiency caused by obstructed activity of natural enemy insects. The random forest algorithm analyzes historical release effect data to optimize the release density under different terrain conditions. In areas with a slope greater than 15 degrees, the release density is increased by 20% to ensure control effectiveness.

[0036] The depth and width parameters of the irrigation ditch were analyzed, and a topographic adaptability assessment was conducted on the ecological barrier point map of the ditch to delineate the boundary of the construction area. Based on the boundary map of the construction area and soil permeability test data, the water storage capacity threshold of the ditch was calculated, and a cross-sectional structure design scheme and irrigation replenishment strategy for the ditch were generated.

[0037] High-resolution topographic images of the ditch construction area are acquired. Ground elevation and slope values ​​are calculated based on the topographic images, and the ditch construction boundary is determined using the slope values. LiDAR point cloud data is collected based on the construction boundary, and a surface model is generated from the point cloud data. Soil profile data is sampled from the surface model to obtain ditch depth and slope ratio parameters. Soil moisture sensors are deployed using the slope ratio parameters, and infiltration data is obtained from the soil moisture sensors. A water transport curve is plotted based on the infiltration data to obtain the ditch's water storage capacity. A ditch cross-section design drawing is generated based on the water storage capacity. An overflow weir and a water level sensor are installed on the ditch cross-section. If the water level sensor detects a water level drop rate exceeding a threshold, a water replenishment device is activated.

[0038] Specifically, topographic images of the ditch construction area were acquired using 0.5-meter resolution remote sensing satellites. A raster calculator was used to calculate ground elevation at 5-meter sampling intervals. Slope values ​​were calculated based on the elevation differences between adjacent raster cells. In areas with slopes less than 15 degrees, the centerline of the ditch was drawn, and the construction boundary was extended to both sides by twice the ditch width, generating a digital topographic map of the ditch construction area. A lidar scanner was used to scan the construction area at a point cloud density of 0.2 meters. A high-precision surface model was generated using a point cloud processor. A random forest algorithm was used to sample soil profiles at a depth of 2 meters every 20 meters in the construction area, obtaining soil texture and lithological stratification data. Combined with the topographic slope, parameters for ditch depth and slope ratio were generated. The ditch depth ranged from 0.8 meters to 1.2 meters, and the slope ratio ranged from 1:1 to 1:1.5.

[0039] Soil moisture sensors were deployed at 50-meter intervals along the centerline of the ditch, measuring soil moisture content every 10 centimeters in the 0-1 meter soil layer, recording data hourly. A soil moisture transport curve was plotted using seven consecutive days of infiltration data. Soil permeability was classified based on the curve slope, and the infiltration loss per unit length of the ditch was calculated. The ditch's water storage capacity was determined by combining this data with annual rainfall data. A cross-sectional design drawing of the ditch was generated using hydrodynamic calculation software. The bottom width was 1-1.5 meters, and the cross-sectional shape was trapezoidal. An overflow weir was installed every 100 meters based on the water storage capacity and terrain slope. Water level sensors were installed on both sides of the ditch to monitor the rate of water level change. When the rate of water level drop exceeded 2 centimeters per hour, a water replenishment device was activated to inject water into the ditch to maintain the designed water level. A resolution of 0.5 meters was chosen for remote sensing data processing primarily to meet the accuracy requirements of the ditch design. At this resolution, 1 square meter of ground area contains 4 pixels, sufficient to identify micro-topographic features. The elevation calculation of the 5-meter sampling interval is based on the spatial scale of the ditch construction machinery operation. The elevation difference between adjacent sampling points reflects the degree of terrain undulation. When the slope exceeds 15 degrees, the risk of soil erosion increases significantly, and the difficulty of ditch construction increases.

[0040] The construction boundaries extending to both sides of the ditch centerline take into account the operating space for construction machinery and the requirements for temporary soil stockpiling; the boundary width is typically 2 to 3 times the width of the ditch. LiDAR scanning uses a point cloud density of 0.2 meters, obtaining 25 elevation measurement points on an area of ​​1 square meter, generating a surface model with centimeter-level accuracy. Soil profiles are sampled every 20 meters; a 2-meter depth profile reveals soil stratification and groundwater levels. Soil texture often changes from loose to dense from top to bottom. Actual cases show that the top 0-30 cm layer is mainly sandy loam with high permeability; 30-100 cm is mainly clay loam with moderate permeability; and below 100 cm is mainly clay with low permeability. The slope ratio is closely related to soil type: 1:1.5 for sandy soil and 1:1 for clay soil. Soil moisture sensors are spaced 50 meters apart, with a measurement point every 10 centimeters vertically, forming a dense monitoring network. Monitoring data shows that after rainfall, the soil moisture content in the top 0-30 cm layer rises rapidly, reaching saturation within 4 hours; the moisture content in the 30-60 cm layer rises slowly, stabilizing after 24 hours; and the moisture content in the layer below 60 cm shows no significant change. The soil moisture transport curve shows significant differences in different soil layers, with a steep curve in the surface layer and a gentler curve in the lower layers. The slope of the curve is positively correlated with the soil permeability coefficient.

[0041] The drainage ditch features a trapezoidal cross-section with a bottom width of 1.2 meters, meeting the design flow requirements. Field observations show that under rainfall intensity of 25 mm per hour, the drainage capacity of the 1.2-meter-wide ditch is 0.3 cubic meters per second. Overflow weirs are spaced 100 meters apart to control water flow velocity and prevent erosion of the ditch bed. Water level sensor records show that the daily water level drop is 4 to 6 centimeters on sunny days, mainly affected by evaporation and seepage. When the drop exceeds 2 centimeters, hourly water replenishment is initiated, with automatic control valves regulating the replenishment flow to maintain the water level within 10 centimeters above and below the design elevation.

[0042] S104. For the parameters of the number of natural enemy insects released and the coordinates of the release points, the release time and release amount are precisely controlled by executing the program through automatic release equipment. Based on the water storage capacity threshold of the ditch and the aquatic ecosystem parameter data, the operation status of the ditch ecological barrier is monitored, the irrigation water replenishment strategy is adjusted in real time, and the ecological barrier effectiveness assessment value is recorded.

[0043] A zigzag cruising route is generated based on the coordinates of the insect predator release points. This route is output by the UAV planner to the automatic release device. Point images are received from a high-speed camera, and insect density data per unit area is obtained using a convolutional neural network. The survival rate is calculated based on this insect density data. Water level data from a water level monitoring sensor is acquired. If the water level is lower than the design level, a water replenishment operation is performed via a solenoid valve. Soil moisture content data on both sides of the ditch is obtained using a soil moisture sensor network. The barrier rate is derived by analyzing the moisture content data and the insect density data using a support vector machine.

[0044] Specifically, a zigzag cruising route is generated based on the coordinates of the predator insect release points and the shortest path planner of the drone. An automatic release device with an accuracy of 0.1 grams is used to release insects at a single point. A weather monitor, sampling every 10 minutes, records the temperature and relative humidity of the release area. A 24-hour activity curve is retrieved from an insect activity database. Release operations are performed when the temperature is between 18 and 25 degrees Celsius and the relative humidity is greater than 75%. A high-speed camera captures images of the release points four times per hour. A convolutional neural network identifies the number of individual predator insects and records the insect density per unit area. The 24-hour survival rate is calculated based on the insect position coordinates in two adjacent images. When the survival rate is below 85%, the release quantity is increased by 15% from the original setting, generating new release parameters for the new points.

[0045] Water level data in the ditch is collected every 30 minutes using a water level monitoring sensor. Leakage and evaporation are recorded using a flow monitor. The range of fluctuations is determined based on monthly curves in the water level change database. When the water level drops by more than 2 cm per hour or falls below 90% of the design level, a solenoid valve is activated to replenish water, with the replenishment flow rate adjusted at 0.5 cubic meters per minute. A soil moisture sensor network is used, with sensors deployed at 10-meter intervals on both sides of the ditch to measure soil moisture content at depths of 0 to 50 cm. The correlation between moisture content change curves and insect density change curves is calculated using a support vector machine. The ecological barrier's blocking effect is calculated from the ditch leakage and pest dispersal rate, generating barrier rate data on a 24-hour cycle. The barrier rate value represents the percentage of pests blocked out of the total pest population.

[0046] The Z-shaped cruising route design for the release points of natural enemy insects is based on the pest dispersal patterns. The distance between adjacent release points is 5 meters, and the drone's flight altitude is maintained at 3 meters to ensure release accuracy. The automatic release equipment uses a volumetric feeder, with a release quantity error controlled within 0.1 grams. Field data shows that under conditions of 20 degrees Celsius and 80% relative humidity, the survival rate of natural enemy insects can reach 95% within 24 hours. However, when the temperature exceeds 28 degrees Celsius or the relative humidity is below 65%, the survival rate drops to below 70%. A high-speed camera with 2 megapixels and a shooting speed of 120 frames per second, equipped with an infrared supplemental light, enables all-weather monitoring. Image recognition uses a tag-re-identification algorithm to number and track individual insects, achieving an accuracy rate of 92%. Under suitable temperature and humidity conditions, the area of ​​a single release point is 78.5 square meters, with an average insect density maintained at 4 to 6 insects per square meter, and the insects are distributed in a circular pattern. When the survival rate decreases, the release quantity is promptly increased to maintain the control density, and the insect density recovers to the standard level within 4 hours after the replenishment.

[0047] The water level monitoring system uses ultrasonic sensors with a measurement accuracy of 1 mm. During the rainy season, when water levels fluctuate frequently, the monitoring frequency is increased to once every 15 minutes. Seepage is measured using cross-sectional flow meters installed at each overflow weir to record hourly flow changes. Actual measurement data shows that normal evaporation loss is 4 mm per day, and soil seepage loss is 8 mm per day. The water replenishment equipment uses variable frequency pumps that automatically adjust the flow rate according to the rate of water level decline, ensuring that water level fluctuations remain within 10 cm above or below the design water level. The soil moisture sensing network uses frequency domain reflectometer sensors, with a measuring point every 10 cm in the 0-50 cm soil layer, forming a 5-layer three-dimensional monitoring network.

[0048] Data shows that when the topsoil moisture content is maintained between 28% and 32%, the pest spread rate decreases to 0.8 meters per day. The barrier rate is calculated based on the difference in pest density on both sides of the ditch. Control sampling points were set up to record pest density under natural spread conditions. A barrier rate above 80% indicates that the ditch's ecological barrier plays a significant role. Monitoring data also shows a significant negative correlation between soil moisture content and pest density; for every 5 percentage point increase in moisture content, pest density decreases by approximately 25%. Through continuous monitoring and timely water replenishment, the ditch's ecological barrier maintains a stable barrier effect.

[0049] S105. Collect environmental parameters of the Dendrobium cultivation area in real time, including temperature, humidity and light data, analyze the impact of environmental parameters on pests and diseases, predict the probability and severity of pests and diseases, and provide a basis for decision-making for the implementation of subsequent prevention and control measures.

[0050] It should be noted that gradient boosting trees are an ensemble learning method that iteratively constructs multiple decision trees and sums the predictions of each tree as the final output. Each new tree is fitted along the negative gradient direction of the prediction residual of the previous tree to gradually reduce the overall prediction error. Its architecture includes a logarithmic loss function to evaluate classification bias, uses a CART regression tree with finite depth as the base learner, and internal nodes split based on thresholds of environmental features (temperature, humidity, light intensity). Leaf nodes output predicted values. The outputs of all decision trees are weighted by the learning rate and summed, then mapped using a logistic function to obtain the probability of pest and disease occurrence, thus outputting a risk prediction result for each monitoring grid unit.

[0051] Temperature, relative humidity, and light intensity parameters are obtained based on the monitoring point number. Outlier detection is performed on the temperature and relative humidity parameters using the three-standard-deviation principle to obtain valid environmental parameters. For these valid environmental parameters, the Dendrobium cultivation area is divided into standard grid units. The influence weights of these valid environmental parameters on pest and disease occurrence are calculated using a random forest algorithm to obtain environmental feature combinations. For these environmental feature combinations, a data clusterer is used to classify the risk levels of the standard grid units. A gradient boosting tree is then used to predict the pest and disease risk of these environmental feature combinations to obtain risk prediction results. If the pest and disease incidence rate within a standard grid unit exceeds a threshold, a risk warning command is triggered. The risk prediction results are then corrected based on the pest and disease incidence rate to obtain a warning level identifier. The pest and disease risk prediction value is calculated using the following formula: in, The combination of environmental features selected by random forest includes six dimensions: temperature (hourly mean, maximum, minimum), relative humidity (mean, standard deviation), and light intensity (cumulative). For the first The predicted values ​​of each CART regression tree (each tree outputs a real value); The total number of decision trees (100 in this example); The learning rate is 0.08 in this embodiment. It is an exponential function. The final output P∈[0,1] represents the probability of pests and diseases occurring within the next 24 hours. A red alert is triggered when P>0.6.

[0052] Specifically, temperature, relative humidity, and light intensity data were collected from environmental monitoring points spaced 10 meters apart. Temperature sensors with 0.1 degrees Celsius accuracy, humidity sensors with 1% accuracy, and light sensors with 10 lux accuracy were used. Environmental parameters were recorded every 10 minutes. Outliers in the temperature and humidity data were identified using a 3x standard deviation principle. An environmental parameter distribution map was generated based on the monitoring point numbers and geographical coordinates. Pest and disease occurrence records were retrieved from the database. Based on the environmental parameter distribution map, the Dendrobium cultivation area was divided into standard 20m x 20m grid units. Environmental parameters within each grid unit were aggregated on an hourly scale. A random forest algorithm was used to calculate the influence weight of environmental parameters on pest and disease occurrence. When the temperature deviation exceeded 2 degrees Celsius, the relative humidity deviation exceeded 5%, or the light intensity deviation exceeded 1000 lux, the combination of environmental parameters for that grid and the corresponding degree of pest and disease occurrence were recorded.

[0053] A data clusterer was used to classify grid cells into five risk levels based on temperature and humidity combinations. Environmental parameter feature combinations were extracted from the clustering results. A gradient boosting tree was used to predict pest and disease risks based on a 30-day environmental parameter sequence, generating a correlation table containing grid number, environmental parameter combination, and pest and disease risk level. The risk level was then adjusted according to the pest and disease incidence rate of individual grid cells. A parameter validator was used to calculate the consistency between the prediction results and actual occurrences. A 7-day sliding time window was used to statistically analyze the prediction accuracy. A risk warning was triggered when the pest and disease incidence rate in a single grid cell exceeded 15%. Four warning levels were established based on the pest and disease incidence rate. Risk areas were marked on the environmental parameter distribution map according to the warning level, generating a prediction result map labeled with the risk level.

[0054] The deployment density of the environmental monitoring network was determined based on the spread characteristics of Dendrobium pests and diseases. Monitoring points spaced 10 meters apart accurately capture microenvironmental changes. Temperature sensors with an accuracy of 0.1 degrees Celsius meet the monitoring requirements for optimal Dendrobium growth temperatures; within the 20-25 degree Celsius range, the incidence of pests and diseases increases by 8% for every 0.5 degree Celsius increase in temperature. Relative humidity with a measurement accuracy of 1% addresses condensation on Dendrobium leaf surfaces; when humidity exceeds 85% and persists for more than 4 hours, the probability of pathogen spore germination significantly increases. Light sensors with an accuracy of 10 lux monitor shading levels; the risk of pests and diseases increases when light intensity is below 8000 lux. Outlier detection uses a 3-standard-deviation principle. Actual measurements show that under normal weather conditions, the temperature variation range at a single monitoring point within 24 hours is within 12 degrees Celsius, and the humidity variation range is within 30%. The 20-meter by 20-meter grid unit division considers plant distribution density and agricultural operation units, with each grid containing approximately 400 Dendrobium plants. The hourly aggregation of environmental parameters used the arithmetic mean, with weighted results showing temperature having a weight of 0.4, humidity 0.35, and light intensity 0.25. Risk levels were classified based on historical pest and disease occurrence data: Level 1 indicated an incidence rate below 5%, Level 2 5% to 10%, Level 3 10% to 15%, Level 4 15% to 20%, and Level 5 above 20%. The combined temperature and humidity characteristics included four statistics: 24-hour average, maximum, minimum, and standard deviation. Light intensity characteristics included sunshine duration and cumulative light intensity. Actual observations showed that when the daily average temperature was 25 degrees Celsius, relative humidity was 88%, and light intensity was 6000 lux, the pest and disease incidence rate reached 18%. The prediction results were validated using a 7-day sliding window to calculate the degree of match between the predicted and actual occurrence levels. Validation data showed that the temperature prediction accuracy reached 85%, the humidity prediction accuracy was 82%, and the combined prediction accuracy reached 88%.

[0055] In the early warning level classification, a yellow alert corresponds to an incidence rate of 15% to 20%, an orange alert to 20% to 25%, a red alert to 25% to 30%, and a purple alert to over 30%. Practice in Dendrobium orchid parks has shown that when there are three consecutive days of red alerts, the probability of a large-scale outbreak of pests and diseases exceeds 90%. At this time, the timeliness of irrigation and ecological barrier control measures directly affects the control effect.

[0056] S106. Based on three parameters—temperature, humidity, and light intensity—generate pest and disease risk warning information for the future period, and identify areas where pests and diseases may be prevalent. When the warning level reaches the threshold, initiate corresponding prevention and control measures, adjust environmental parameters, and strengthen real-time monitoring.

[0057] Three parameters—temperature, humidity, and light—are obtained based on an environmental factor association model. A recurrent neural network is used to perform sequence prediction on these parameters to obtain prediction results. Based on the prediction results, a pest and disease risk value is calculated. If the pest and disease risk value within a grid cell exceeds a warning threshold, the environmental parameters of that grid cell are automatically adjusted. Shading and ventilation devices are used to adjust the environmental parameters within the grid cell. Feedback data from within the grid cell is collected, and closed-loop control of temperature and humidity values ​​is implemented. Images of Dendrobium plants within the grid cell are acquired using an image acquisition device. A convolutional neural network is used to identify pest and disease characteristics in the images to obtain the lesion coverage rate. The frequency of environmental parameter adjustment is adjusted based on the lesion coverage rate. The adjustment frequency refers to the execution frequency of environmental parameter regulation, for example, adjusting from once every 30 minutes to once every 15 minutes to respond more quickly to pest and disease development.

[0058] Specifically, the system reads three prediction parameters—temperature, humidity, and light intensity—and uses a recurrent neural network to sequentially predict environmental parameters for the next 7 days at 1-hour intervals. Based on the prediction results, it calculates pest and disease risk values ​​and divides the cultivation area into standard 20m x 20m grids. When the 24-hour pest and disease risk value within a grid cell exceeds 0.8, that grid cell is marked with a red warning level, generating a warning area distribution map. A 5-point weighted average filter is used to smooth the environmental parameter prediction results. The pest and disease spread rate is calculated based on the temperature and humidity gradient between adjacent grid cells. Risk level thresholds are extracted from the warning rule base: 0.8 for red, 0.6 for orange, and 0.4 for yellow. If the risk value of a single grid cell exceeds the warning threshold for 4 consecutive hours, automatic adjustment of the environmental parameters for that grid cell is initiated. The environmental index range of the early warning grid unit is read from the parameter control database. The temperature is controlled between 18 and 25 degrees Celsius, the relative humidity is controlled between 65% and 80%, and the light intensity is controlled between 5,000 and 15,000 lux. The environmental parameters are adjusted through shading and ventilation devices. Feedback data is collected every 30 minutes to perform closed-loop control of the temperature and humidity values ​​in the early warning grid unit.

[0059] Image acquisition points were set up at 5-meter intervals within the early warning grid units. A 2-megapixel camera captured images of Dendrobium plants once per hour. A convolutional neural network was used to identify pest and disease features in the images, calculating the number and area of ​​lesions per unit leaf area. When the lesion coverage exceeded 5% of the leaf area, the environmental parameter control was strengthened, increasing the sampling frequency to once every 15 minutes. In the environmental parameter prediction, the selection of temperature, humidity, and light was based on the physiological characteristics of Dendrobium. Temperature showed an exponential relationship with the pathogen's reproduction rate, with 25 degrees Celsius being a critical turning point; above this temperature, the pathogen's reproduction rate increased significantly. The 1-hour prediction interval corresponded to the pathogen's spore germination cycle. Actual data showed that under suitable conditions, it takes 2 to 3 hours for spores to go from germination to invasion. The 20-meter grid division was based on the pathogen's spore diffusion radius; the diffusion range of a single spore within 24 hours typically does not exceed 15 meters.

[0060] Environmental parameter smoothing employed a 5-point weighted average method to eliminate short-term fluctuations, with weights of 0.1, 0.2, 0.4, 0.2, and 0.1. The spread rate of pests and diseases was closely related to the temperature and humidity gradient; for every 1 degree Celsius increase in temperature difference between adjacent grids, the spread rate increased by 20%. The warning level was set with reference to the pathogen's reproductive cycle; 4 hours corresponded to the complete process from spore germination to invasion, and a risk value of 0.8 indicated an 80% probability of pathogen reproduction under these environmental conditions. The determination of the environmental parameter control range comprehensively considered both the growth needs of Dendrobium and the requirements for pathogen suppression. A temperature range of 18 to 25 degrees Celsius satisfied both Dendrobium photosynthesis and inhibited pathogen reproduction. Relative humidity was controlled between 65% and 80%, below the 85% humidity threshold required for pathogen spore germination. Light intensity of 5000 to 15000 lux ensured photosynthetic efficiency while simultaneously inhibiting pathogen activity through ultraviolet radiation. The 30-minute adjustment cycle of the shading and ventilation devices corresponded to the inertia of environmental parameter changes.

[0061] Image sampling density is based on the lesion expansion rate, with a 5-meter interval enabling timely detection of early-stage lesions. At 2-megapixel resolution, each pixel corresponds to an actual area of ​​0.1 square millimeters, allowing the identification of early lesions as small as 0.5 millimeters in diameter. A lesion area coverage of 5% is the critical point for disease control; exceeding this percentage leads to a sharp increase in disease expansion rate. Image sampling frequency increases with disease severity, with a 15-minute sampling interval used to track lesion expansion rate and provide real-time feedback for environmental parameter adjustment. Practice shows that timely detection and control of environmental conditions can control the disease in its early stages, limiting lesion expansion to within 2% of the leaf area.

[0062] S107. Based on historical data of the Dendrobium cultivation area pre-input by the UAV and current image data of the monitored Dendrobium, a comprehensive assessment of the Dendrobium growth status is conducted using time series analysis to determine if any abnormalities exist. If growth data is detected to deviate from the normal range multiple times consecutively, an alert is issued indicating potential pest or disease problems affecting Dendrobium growth, prompting management personnel to conduct on-site verification and handling.

[0063] It should be noted that time series analysis refers to arranging growth indicators of Dendrobium plants, such as plant height, number of leaves, and chlorophyll content, in chronological order of collection time to form time series data. A recurrent neural network (specifically a Long Short-Term Memory network, LSTM) is used to model the sequence and fit a growth curve. The LSTM architecture includes a forget gate, an input gate, and an output gate, enabling it to learn long-term dependencies. The input is the growth indicator sequence of the past 7 days, and the output is the predicted value for the next 24 hours. By comparing the predicted value with the normal growth range, it is determined whether the growth indicators deviate continuously, thus issuing an early warning.

[0064] Images of Dendrobium plants captured by a drone camera are acquired. An image processor measures plant growth parameters, including plant height, number of leaves, and chlorophyll content. Based on these parameters, a recurrent neural network is used to fit a growth curve, yielding the daily growth rate of plant height, leaf unfolding speed, and chlorophyll content change rate. An image recognition device compares the plant images with pest and disease maps for feature similarity. If the feature similarity score exceeds a threshold, a pest and disease risk record containing the plant number and feature type is generated. A long short-term memory network is used to perform time-series analysis on the pest and disease risk record. If plant growth indicators remain at a preset deviation level and the feature similarity score exceeds a threshold, a warning message is sent to the data terminal.

[0065] Specifically, images of Dendrobium plants were captured every two hours using a 2-megapixel drone camera. An image processor measured plant height, number of leaves, leaf area, and chlorophyll content. Growth parameters were recorded according to plot number and collection timestamp. A recurrent neural network was used to fit the plant image features to historical growth data, generating a 30-day growth curve. Based on the growth curve, the daily growth rate of plant height, leaf unfolding speed, and chlorophyll content change rate were calculated. Values ​​within the normal growth range for the same period were extracted from a growth parameter database: daily plant height growth rate of 0.2 to 0.5 cm, leaf unfolding speed of 2 to 4 leaves per week, and relative chlorophyll content of 35 to 45 units. When plant indicators fell below the lower limit of the normal range, an anomaly marker was recorded, and the degree of anomaly was classified into four levels. Pest and disease images are obtained from the feature database and classified into four categories: leaf discoloration, curling, wilting, and malformation. An image recognition device is used to compare the abnormal features of the plant with the pest and disease images and calculate the feature similarity score. When the similarity exceeds 0.85, it is judged as a suspected pest or disease and generates a pest and disease risk record containing the plant number, feature type, and similarity score.

[0066] A long short-term memory (LSTM) network was used to perform time-series analysis on 7 consecutive days of pest and disease risk records. Historical occurrence patterns were retrieved from the pest and disease database. When plant growth indicators remained at deviation level 3 or above for 72 consecutive hours and the feature similarity score exceeded 0.85, a red alert signal was triggered, sending an alert message to the data terminal containing the plant number, abnormal feature type, and risk level. Dendrobium growth monitoring employed a 2-megapixel camera, with each pixel corresponding to an actual area of ​​0.1 square millimeters. At a 2-meter aerial photography height, it could accurately identify leaf lesions as small as 0.5 millimeters. The 2-hour data collection interval was based on the growth characteristics of Dendrobium leaves; changes in stem and leaf growth became observable within 4 to 6 hours. Consistency of image data was ensured by setting fixed aerial photography positions and angles.

[0067] Image processing employed a multi-scale segmentation method, controlling plant height measurement error to within 0.1 cm and achieving a leaf counting accuracy of 98%. Normal ranges for growth parameters were determined through statistical analysis. Taking stem growth as an example, during the suitable growing season, the daily growth rate of healthy plants remained between 0.3 and 0.4 cm; a rate below 0.2 cm for three consecutive days indicated stunted growth. Leaf unfolding speed was affected by temperature and light conditions; under conditions of 20-25 degrees Celsius and 70-80% relative humidity, 2-3 new leaves unfolded per week. Chlorophyll relative content was measured using the chlorophyll fluorescence method; healthy leaves showed a value of around 40 units, while the rapidc value dropped below 30 units when infected with disease.

[0068] The identification of pest and disease characteristics focuses on changes in leaf morphology. For example, thrips damage causes silvery-white streaks on affected leaves, with the leaves curling inwards; the similarity to these characteristics exceeds 0.9. Anthracnose causes circular brown spots on leaves with distinct edges, and the lesions are 2 to 8 mm in diameter. Soft rot causes water-soaked rot in leaf tissue, presenting as irregular brown spots. When a disease characteristic is identified and the similarity exceeds 0.85, it is considered a suspected case of that disease. The time-series analysis uses a 72-hour observation window, primarily considering the disease development cycle. For example, for leaf spot disease, it takes approximately 48 to 60 hours from pathogen invasion to symptom appearance. In the risk level classification, Level 1 indicates a slight deviation from normal growth indicators; Level 2 indicates the appearance of typical disease characteristics; Level 3 indicates an expansion of the disease area or symptoms appearing in multiple locations; and Level 4 indicates an accelerated disease spread and increased severity. Practice shows that when a plant is in a risk state of Level 3 or higher for 72 consecutive hours, untreated disease will rapidly spread to surrounding plants.

[0069] S108. Continuously optimize the ecological environment of the Dendrobium cultivation area. Through regular cleaning of ditches and water quality monitoring, a natural physical barrier is formed to prevent underground pests from invading. At the same time, through an automated temperature and humidity control system, environmental parameters are adjusted in real time according to climate changes to always maintain the optimal environmental conditions for Dendrobium growth, thus controlling the occurrence of pests and diseases at the source.

[0070] Water quality data, including dissolved oxygen, pH, and turbidity, is acquired using multi-parameter water quality sensors. Based on this data, it is determined whether to activate the circulating filtration pump and add flocculant. Soil moisture content data on both sides of the ditch is acquired through a soil moisture sensor network. A random forest algorithm is used to determine the relationship between soil moisture content and infiltration rate. Based on the infiltration rate, it is determined whether to activate the water replenishment equipment. Environmental parameters of the cultivation area are acquired using temperature and humidity sensors. Based on these parameters, it is determined whether to activate heating and dehumidification equipment. A convolutional neural network is used to perform spatial analysis on the measurement point data within the cultivation area. Based on the analysis results, control parameters for ventilation and shading devices are determined, and the equipment operation is adjusted accordingly.

[0071] It should be noted that the ventilation devices involved in this method are not designed for the ventilation needs of enclosed greenhouses, but rather for the active regulation of the local microenvironment within the plant canopy in open-field Dendrobium cultivation. In open-field environments, although overall air circulation is good, the high planting density of Dendrobium plants and the overlapping leaves make it difficult for natural wind to effectively penetrate the canopy. This leads to prolonged water retention on the leaf surface after rain or irrigation, creating a localized high-temperature and high-humidity microenvironment, which easily induces fungal diseases such as anthracnose and leaf spot. The ventilation devices used (such as mobile axial flow fans or suspended circulating fans) actively force airflow, accelerating the evaporation of water from the leaf surface and disrupting the high-humidity conditions (relative humidity >85% for more than 4 hours) required for pathogen spore germination. Simultaneously, they are linked to shading devices to compensate for natural ventilation obstruction caused by the deployment of shading nets, thereby achieving precise environmental control under open-field cultivation conditions and reducing the risk of pests and diseases.

[0072] Specifically, a multi-parameter water quality sensor collects dissolved oxygen, pH, and turbidity data in the ditch every 30 minutes. If dissolved oxygen falls below 5 mg / L or turbidity exceeds 50 Niffler units, the circulating filtration pump automatically activates to remove floating debris. Flocculant is added with an accuracy of 0.1 mg / L based on the turbidity value. The water purification process is monitored in real time using a water quality analyzer. When the water quality indicators return to normal, the amount of cleaning agent used and the purification time are recorded. A soil moisture sensor network is used to deploy monitoring points at 5-meter intervals on both sides of the ditch. A random forest algorithm is used to calculate the relationship between soil moisture content and ditch infiltration. Based on the infiltration data, the water level is adjusted between 0.8 meters and 1.2 meters. When the infiltration loss exceeds 0.5 cubic meters, the water replenishment equipment is activated every hour to maintain the soil moisture content of the ditch barrier zone between 25% and 30%.

[0073] Based on environmental data collected every 10 minutes by temperature and humidity sensors, and comparing it with the parameter range of 20-25 degrees Celsius and 65%-75% relative humidity required for Dendrobium growth, the operation of heating and dehumidification equipment is adjusted by an automatic controller. When the temperature is below 18 degrees Celsius, the hot air blower is turned on; when the relative humidity exceeds 80%, the dehumidifier is turned on, achieving precise control of environmental parameters in the cultivation area. A convolutional neural network is used to perform spatial analysis of the temperature and humidity distribution within the cultivation area. An environmental parameter distribution map is generated based on temperature and humidity data from 24 measuring points. The adjustment parameters for ventilation and shading equipment are calculated from the temperature difference and humidity gradient between the cultivation area and the outside. Ventilation devices (such as fans) are activated when the temperature difference is greater than 5 degrees Celsius, and shading devices (such as shade nets) are deployed when the light intensity exceeds 15,000 lux, maintaining a dynamic balance of environmental parameters within the cultivation area.

[0074] Dissolved oxygen is a key indicator reflecting the ecological status of aquatic bodies in water quality monitoring. When it falls below 5 mg / L, anaerobic bacteria begin to proliferate rapidly, accelerating water quality deterioration. When water turbidity exceeds 50 Neifer units, the excessive suspended solids content impairs the water body's self-purification capacity. Circulating filtration uses a 0.2 mm pore size filter to remove floating debris. The flocculant dosage is linearly related to turbidity; 0.5 mg / L of flocculant is required for every 100 Neifer units of turbidity. Actual measurements show that the flocculation and sedimentation process takes 45 to 60 minutes, and the turbidity of the water can be reduced to below 10 Neifer units after sedimentation. The soil moisture monitoring network deployment density is based on the lateral migration pattern of water; a 5-meter spacing accurately monitors the water diffusion front.

[0075] The infiltration rate exhibits a non-linear relationship with soil moisture content, significantly increasing when soil moisture content exceeds 30%. Maintaining the water level in the irrigation ditches at approximately 1 meter is most economical, as this balances infiltration loss with replenishment. The soil moisture content in the barrier zone is maintained between 25% and 30%, ensuring root absorption while preventing underground pests from penetrating. Environmental control in the cultivation area employs a zoned adjustment strategy. Temperature sensors are arranged vertically in three layers (top, middle, and bottom), with eight measuring points in each layer forming a monitoring network. Heating equipment utilizes a hot air blower with a rated power of 12 kW and an air volume of 2000 cubic meters per hour, activated when the temperature drops below 18 degrees Celsius, with a heating rate of 2 degrees Celsius per hour. The dehumidifier has a cooling capacity of 10 kW and a dehumidification capacity of 15 liters per hour, activated when the relative humidity exceeds 80%, with a dehumidification rate of 3% per hour. Spatial environmental analysis of the cultivation area, based on data from 24 measuring points, yielded isotherms and isohyets, revealing the distribution patterns of temperature and humidity.

[0076] Actual measurement data shows that when the temperature difference between the cultivation area and the outside exceeds 5 degrees Celsius, the temperature at the top is 2 to 3 degrees Celsius higher than near the ground. At this point, activating the ventilation system can quickly and evenly lower the temperature. Shading control is linked to light intensity; 15,000 lux is the light saturation point. Beyond this value, light energy utilization efficiency decreases. After deploying the shading system, the light transmittance drops to 45%, ensuring photosynthesis while preventing excessive temperature. When the equipment is controlled in a coordinated manner, ventilation is prioritized for cooling, and shading is only used when ventilation is insufficient, forming an energy-efficient and effective control scheme.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for controlling pests and diseases in Dendrobium cultivation, characterized in that, include: The system acquires aerial images taken by drones, identifies suspected pest and disease areas based on these images, and sends the locations of the suspected pest and disease areas to the backend system. The back-end system controls the drone to perform a detailed scan of the suspected pest and disease area, and combines color difference thermal imaging detection to confirm the type and distribution of pests and diseases. Based on the types and distribution of pests and diseases, analyze the migration paths of pests, calculate the release parameters of natural enemy insects, and generate a map of ecological barrier points in ditches using the drone aerial images to obtain a prevention and control plan. According to the prevention and control plan, natural enemy insects are released through automatic release equipment, and the water level in the ditch is monitored by a water level sensor. When the water level is lower than the threshold, water replenishment is initiated, and the barrier effect is evaluated by the barrier rate. Environmental parameters are acquired, their impact weights on pests and diseases are analyzed, and pest and disease risk is predicted to obtain the risk prediction results. Based on the risk prediction results, when the risk value exceeds the warning threshold, control measures are initiated, and plant images are collected to identify the lesion coverage rate. The control frequency is then adjusted based on the lesion coverage rate. The system acquires aerial images of Dendrobium plants taken by drones and uses time series analysis to fit growth curves. When growth indicators continuously deviate from the normal range, it sends early warning information. Monitor the water quality of the ditch and obtain data from the measuring points in the cultivation area. Start the circulation filtration and water replenishment. Perform spatial analysis on the measuring point data in the cultivation area to determine the control parameters of the shading and ventilation devices and maintain the optimal environment for Dendrobium growth.

2. The method for controlling pests and diseases in Dendrobium cultivation according to claim 1, characterized in that, The system acquires multispectral red and near-infrared images from drones, extracts leaf reflectance through denoising and segmentation, calculates the normalized vegetation index as a health index, and locates suspected areas when the health index is below a threshold. These areas are then compared with a spectral database of pest and disease characteristics for confirmation, and the records are sent to the backend system.

3. The method for controlling pests and diseases in Dendrobium cultivation according to claim 1, characterized in that, The backend system controls the drone to perform a detailed scan of the suspected pest and disease area, and combines color difference thermal imaging detection to confirm the type and distribution of pests and diseases, including: The system receives spectral sensor data and thermal imaging sensor data collected by a drone. The data is obtained by scanning the suspected pest and disease area according to a sampling grid. Based on the data, region segmentation is performed to obtain a grayscale histogram of the plant leaf region. If the color difference value of adjacent pixels in the grayscale histogram exceeds the threshold, an abnormal spot map is obtained. Temperature samples were taken from the plant leaves and root areas. The degree of temperature anomaly was calculated using a support vector machine. When the degree of temperature anomaly exceeded a threshold, the area was identified as a suspected disease site. For the suspected disease points, a ResNet network is used to extract feature distribution maps, and similarity matching is performed using standard feature templates in the disease feature library. If the similarity exceeds a preset matching threshold, it is determined that the corresponding disease exists in the area.

4. The method for controlling pests and diseases in Dendrobium cultivation according to claim 1, characterized in that, The process involves analyzing pest migration paths based on the types and distribution of pests, calculating natural enemy insect release parameters, and generating a map of ecological barrier points in the ditch using drone aerial images to arrive at a control plan, including: Based on the data collected from the grid, pest density distribution data is obtained, and a convolutional neural network is used to track the activity trajectory of pests to obtain a pest migration path trajectory map. Soil moisture sensors are used to acquire spatial distribution data of soil permeability, and the location of the barrier zone is determined based on the spatial overlap between the pest migration path trajectory map and the spatial distribution data of soil permeability. The monitoring units are divided according to the location of the barrier zone. The insect density of the monitoring units is statistically analyzed to obtain the risk level. The number of natural enemy insects to be released in each monitoring unit is determined based on the predation parameters of natural enemy insects and the risk level.

5. The method for controlling pests and diseases in Dendrobium cultivation according to claim 4, characterized in that, If the monitoring unit has the number of predator insects released, the random forest algorithm is used to optimize the spatial distribution of the number of predator insects, the coordinates of the release point are calculated based on the predation radius, and the number of predators at the release point is determined according to the risk level of the unit.

6. The method for controlling pests and diseases in Dendrobium cultivation according to claim 1, characterized in that, A cruise route is generated based on the release points of the natural enemy insects in the prevention and control plan, and the release of natural enemy insects is carried out by the automatic release equipment; images of the release points are collected to identify the insect density per unit area and calculate the survival rate; the water level of the ditch is monitored by the water level sensor, and water replenishment is initiated when the water level is lower than the threshold; the barrier rate is evaluated by analyzing the soil moisture content and insect density data on both sides of the ditch.

7. The method for controlling pests and diseases in Dendrobium cultivation according to claim 1, characterized in that, The influence weights of each environmental parameter on the occurrence of pests and diseases are calculated using the random forest algorithm to obtain a combination of environmental features. The gradient boosting tree is then used to predict the risk of pests and diseases based on the combination of environmental features, and the risk prediction result is obtained.

8. The method for controlling pests and diseases in Dendrobium cultivation according to claim 1, characterized in that, Based on temperature, humidity, and light parameters, a sequence prediction is performed to calculate the risk value of pests and diseases; when the risk value exceeds the warning threshold, the environmental parameters of the corresponding area are automatically adjusted through the shading and ventilation device.

9. The method for controlling pests and diseases in Dendrobium cultivation according to claim 1, characterized in that, The plant height, number of leaves, and chlorophyll content are measured from the aerial images of Dendrobium plants taken by the UAV. The growth curve is fitted using time series analysis to obtain the daily growth rate of plant height, leaf unfolding speed, and chlorophyll change rate. If the feature similarity exceeds the threshold and the growth indicators continuously deviate from the normal range, the warning information is sent through image recognition and comparison with disease and pest maps.

10. The method for controlling pests and diseases in Dendrobium cultivation according to claim 1, characterized in that, A convolutional neural network was used to perform spatial analysis on the measurement point data in the cultivation area. Based on the analysis results, the control parameters of the shading and ventilation device were determined to maintain the optimal environment for Dendrobium growth.

Images